brycewang-stanford FAccT Review Process
Model the pipeline before interpreting any single review. FAccT's process has two features that surprise authors arriving from a pure-ML venue: the reviewer pool is interdisciplinary (a computer scientist, a lawyer, and a social scientist may all be assigned), and — new for the 2026 edition — the decision set is Accept / Revise / Reject, where Revise is a genuine revise-and-resubmit round, not a soft reject. Your paper is matched to reviewers and Area Chairs by the focus areas you selected at registration, so those choices shape who reads you as much as your title does.
Process model
- Submission and review run on OpenReview (new for 2026) with mutual anonymity: authors and reviewers are hidden from each other.
- Papers are matched to reviewers and Area Chairs by disciplinary focus area, so an interdisciplinary paper is typically read by people from more than one field — a strength and a risk (each expects their lane's rigor).
- Reviewers weigh relevance to the conference and the chosen area, and quality and clarity — correctness, depth of exposition (how well you contextualize approach, methodology, perspective, and domain), and an honest evaluation of both strengths and limitations.
- After preliminary reviews, a short rebuttal window lets authors correct factual errors and misunderstandings and signpost minor edits — it is not for re-arguing substantive disagreement.
- Area Chairs synthesize reviews into a recommendation: Accept, Revise, or Reject. Revise papers address AC-prioritized concerns by the revision deadline and are re-reviewed before a final decision.
- Accepted papers are archival ACM proceedings (or non-archival by author choice) and are presented in person.
Reading a decision against the categories
| Decision | What it means | Author move |
|---|---|---|
| Accept | Contribution and rigor hold; minor polish only | Camera-ready; do not reopen scope |
| Revise | Repairable gaps: a missing disaggregation, an under-argued harm, a construct concern, a thin cross-lane engagement | Treat as an R&R: address each AC-prioritized concern, evidenced, by the deadline |
| Reject | Structural: not FAccT-shaped, harm claim unsupported, one-discipline paper misrouted | Reframe or reroute (NeurIPS/ICML/CHI/AIES/law), do not lightly resubmit unchanged |
The strategic reading: write the initial submission so that whatever is weakest is fixable inside the revision window (an analysis you can add, a limitation you can bound, a construct you can validate) rather than structural (a study design or a venue-fit problem you cannot repair in weeks). The 2026 process is built to reward repairable papers.
The interdisciplinary reviewer reality
Expect readers from different fields, matched by your focus areas. A fairness-methods reviewer checks your metrics, baselines, and disaggregation; a legal reviewer checks whether you use the doctrine correctly; a qualitative reviewer checks your coding, reflexivity, and treatment of participants. The common failure is a paper strong for one and naive to another — so answer each reviewer on the axis they raised, and do not dismiss a cross-lane objection as "not my field": at FAccT it is precisely the point.
Where author leverage actually exists
[Before submission] focus-area selection -> which disciplines review you (largest lever)
[Rebuttal] factual corrections and misreadings only; not a debate
[Revise round] the strongest lever: address each AC-prioritized concern with concrete changes,
re-read before the final decision
[After reject] no appeal; reroute to a sibling venue or the CRAFT track, or resubmit next cycle
A rebuttal moves borderline papers when it fixes a factual misreading a reviewer built an objection on; it does not move papers when it argues taste. In the Revise round, an unaddressed AC-prioritized concern — neither fixed nor explicitly and reasonably declined — is what turns the re-review against you.
Reading a review packet
Weight reviews before answering. A review that engages your subgroup tables, your codebook, or the specific doctrine was read closely and will be read closely again — that reviewer is your likely advocate if the revision holds. A review from an adjacent discipline that raises a framing or harm concern is not noise; it is the interdisciplinary check the venue exists for, and the Area Chair will weight it.
Misreadings to avoid
- Treating Revise as a guaranteed accept — the re-review is real; budget the short window like a deadline.
- Treating the rebuttal as the revision — the rebuttal is for factual corrections; substantive fixes belong in the Revise round.
- Dismissing the out-of-field reviewer — the mixed panel is the design, not an accident.
- Projecting a pure-ML process — FAccT is not single-shot accept/reject; and the Revise round, OpenReview, and rebuttal specifics are 2026-new, so re-confirm them each cycle.
Output format
[Process stage] pre-submission / awaiting reviews / rebuttal / revise / final / accepted
[Decision category] accept / revise / reject, with the criterion driving it
[Criterion map] each review point -> relevance | correctness | depth | strengths-and-limits | discipline
[Leverage plan] the next-stage action that can actually change the outcome
[Forbidden moves] identity leak / arguing taste in rebuttal / ignoring the out-of-field reviewer